Hypoxia Inducible Factor-2α Regulates the Development of Retinal Astrocytic Network by Maintaining Adequate Supply of Astrocyte Progenitors

Here we investigate the role of hypoxia inducible factor (HIF)-2α in coordinating the development of retinal astrocytic and vascular networks. Three Cre mouse lines were used to disrupt floxed Hif-2α, including Rosa26CreERT2, Tie2Cre, and GFAPCre. Global Hif-2α disruption by Rosa26CreERT2 led to reduced astrocytic and vascular development in neonatal retinas, whereas endothelial disruption by Tie2Cre had no apparent effects. Hif-2α deletion in astrocyte progenitors by GFAPCre significantly interfered with the development of astrocytic networks, which failed to reach the retinal periphery and were incapable of supporting vascular development. Perplexingly, the abundance of strongly GFAP+ mature astrocytes transiently increased at P0 before they began to lag behind the normal controls by P3. Pax2+ and PDGFRα+ astrocytic progenitors and immature astrocytes were dramatically diminished at all stages examined. Despite decreased number of astrocyte progenitors, their proliferation index or apoptosis was not altered. The above data can be reconciled by proposing that HIF-2α is required for maintaining the supply of astrocyte progenitors by slowing down their differentiation into non-proliferative mature astrocytes. HIF-2α deficiency in astrocyte progenitors may accelerate their differentiation into astrocytes, a change which greatly interferes with the replenishment of astrocyte progenitors due to insufficient time for proliferation. Rapidly declining progenitor supply may lead to premature cessation of astrocyte development. Given that HIF-2α protein undergoes oxygen dependent degradation, an interesting possibility is that retinal blood vessels may regulate astrocyte differentiation through their oxygen delivery function. While our findings support the consensus that retinal astrocytic template guides vascular development, they also raise the possibility that astrocytic and vascular networks may mutually regulate each other's development, mediated at least in part by HIF-2α.


Introduction
Retinal vascular development is closely associated with the development of an astrocytic template. Earlier studies found that retinal astrocytes were present in animal species with vascularized retinas but absent in animals with avascular retinas [1]. More recent studies employed gene targeting approach in mice to address the relationship between astrocytic and vascular development. Tlx null mutation led to poor astrocytic and vascular development in the retina [2]. Besides reduced astrocyte numbers, Tlx null mice also displayed poor assembly of extracellular fibronectin matrices [3], and astrocyte specific Tlx disruption further demonstrated that the expression of both fibronectin and heparan-sulfate was compromised [4]. These extracellular components were thought to mediate retinal vascularization by regulating VEGF-A binding and distribution [4].
In rodents, the development of retinal astrocyte network begins at birth with immigration of Pax2-positive cells from the optic nerve, spreading in a centrifugal direction in the retinal inner surface [5,6]. Pax2 + cell population gives rise to both optic nerve astrocytes and retinal astrocytes, with the progenitors to the latter also expressing PDGFRa in addition to Pax2 [7]. PDGFRa expression is critical to the proliferation of immature retinal astrocytes in response to stimulation by PDGFA from retinal ganglion cells [8]. As astrocyte progenitors migrate towards the retinal periphery, vascular structures emerge from the optic nerve, forming a vascular network which expands towards retinal periphery behind the PDGFRa + astrocytic network. In vascularized areas, astrocyte maturation occurs, presumably mediated by endothelial cell derived leukemia inhibitory factor (LIF) [9][10][11]. Mature astrocytes exhibit high level expression of glial fibrillary acidic protein (GFAP), whereas Pax2 expression is lost [12][13][14].
How the astrocytic network facilitates retinal vascular development remains incompletely understood. In spite of the important role of VEGF-A in vascular development [15][16][17], astrocyte specific disruption of VEGF-A expression did not interfere with retinal vascular growth, although vascular stability was compromised [18]. Thus, VEGF-A for retinal vascular development is presumably derived from non-astrocytic cells. However, VEGF-A extracellular distribution in developing retinas may be controlled by astrocyte-derived fibronectin and heparan sulfate [3,4]. In addition, R-cadherin in retinal astrocytes is also important for retinal vascular development, which was demonstrated by blocking R-cadherin function with a neutralizing antibody [19].
A recent study found that HIF-1a deficiency in retinal neural tissues led to compromised development of both astrocytic and vascular networks [20]. In the present study, we compare contributions of HIF-1a and HIF-2a to the development of retinal astrocytic and vascular networks, with an emphasis on the role of HIF-2a in the astrocytic lineage. Selective Hif-2a disruption in Pax2 + astrocyte progenitor cells led to precocious and accelerated differentiation of Pax2 + progenitors into GFAP + astrocytes, causing a shortage in the supply of Pax2 + progenitors and premature cessation of astrocyte development. Because HIF-2a protein undergoes oxygen dependent degradation, our findings suggest that retinal vascular development may modulate astrocyte development by regulating Hif-2a protein levels.

Mice
All animal procedures were approved by the Animal Care Committee at the University of Connecticut Health Center in compliance with Animal Welfare Assurance. Mice were housed with a 12 light/12 darkness cycle, and were maintained on normal chow. Mice were bred by natural mating, and the day when a litter was born was designated P0.
Floxed Hif-1a mice were originally produced by Randall Johnson's lab [21] and purchased from the Jackson laboratory in C57BL/6 (B6 from here on) strain background. These mice were crossed with CD1 and subsequently maintained in B6/CD1 mixed background. Floxed Hif-2a mice were produced in our own lab from B6/129 hybrid ES cells [22], and then backcrossed to B6 for 4 generations. At this point reproduction slowed down significantly, and a subgroup of the mice were crossed with CD1 females, leading to a population of floxed Hif-2a mice in mixed CD1 and B6 background at approximately 50%:50% ratio. Two GFAP Cre lines were purchased from the Jackson laboratory, including a line originally generated by Albee Messing (Jax stock number 004600) [23] and another line donated by Michael Sofroniew (line 77.6, Jax stock number 012887) [24]. GFAP Cre mice originating from the Messing lab were supplied in FVB/N background. These mice were backcrossed into B6 for 4 generations, before they were crossed with CD1 females, resulting in mixed CD1/B6 background similar to that in floxed Hif-2a mice. The line 77.6 GFAP Cre mice from the Sofroniew lab were received in B6 strain background, and were crossed into CD1 by one generation. The tdTomato mice carried a CAG promoter-loxP-Stop-loxP-tdTomato transgene targeted into the ubiquitously expressed Rosa26 locus [25], and were supplied in B6 strain background. These mice were crossed to CD1 by one generation before being crossed to GFAP Cre mice. Tie2 Cre was originally generated by Richard Flavell's group [26], and supplied by the Jackson lab (Jax stock number 004128). Detailed breeding information is summarized in Table S1.
Cre recombinases in Tie2 Cre or GFAP Cre mice were both constitutively active. The Rosa26 CreERT2 mouse line was a gift from A. Joyner (New York University School of Medicine, New York, New York, USA) and was similar to a related mouse line described by Shebler et. al. [27,28]. Rosa26 CreERT2 -encoded CreERT2 was activated by tamoxifen. Unless specifically indicated, neonatal mice were treated with tamoxifen by daily oral gavage from postnatal day 1 (P1) through P3 (40 mg/kg of body mass, with tamoxifen dissolved in corn oil). In some experiments, pregnant mothers were treated with tamoxifen by oral gavage at 17.5 days post coitum, followed by two more doses delivered to neonates at P1 and P2. In addition, some mice were treated with tamoxifen at P0 through P2. In all cases, pups with or without Rosa26 CreERT2 were treated with tamoxifen in parallel.
Whole mount staining of neonatal retinas with isolectin B4 conjugated to Alexa fluorH-594(IB 4 -Alexa 594) IB 4 staining of neonatal retinas was carried out as described before [30]. Briefly, neonatal mice were euthanized by decapitation at indicated stages between P0 and P8, and eyes were isolated by enucleation. Isolated eyes were fixed with 4% paraformaldehyde (PFA) for 45 minutes at room temperature, following which retinas were dissected out. Retinas were cut by four incomplete radial incisions, leaving 4 pedals attached to one another at the center. Retinas were incubated at 4uC overnight with IB 4 -Alexa 594 (Life Technologies) at 1:100 dilution in Retina Staining Buffer (RSB), which consisted of phosphate buffered saline (PBS), 1 mM CaCl 2 , 1 mM MgCl 2 , 1% Triton X100, and 1% BSA. Stained retinas were washed three times in RSB (1 hour each with rocking), and flat-mounted in 50% glycerol in PBS. Imaging was carried out by confocal microscopy.

Detection of proliferative and apoptotic astrocyte progenitors
For proliferation assay, neonatal mice were injected with 5bromo-2-deoxyuridine (BrdU, 120 mg/kg of body mass; Roche Applied Science, Indianapolis, IN), and euthanized after 60 minutes. Retinas were fixed in 4% paraformaldehyde with the rest of the eyes still attached, and stored in 70% ethanol overnight. Thereafter, fixed retinas were dissected away from other eye structures, treated with 1% Triton X-100 in PBS for 30 minutes at room temperature, and then incubated in 2 mol/L HCl for 1 hour at 37uC. Processed retinas were double stained with mouse anti-BrdU antibody (BD Biosciences, San Jose, CA) and rabbit anti-Pax2, followed by goat anti-mouse IgG-Alexa fluorH-594 and goat anti-rabbit IgG-Alexa fluorH-488.
Apoptosis was monitored by IF staining with anti-active Caspase 3 (Abcam, Cambridge, MA). For positive control, neonatal mice were exposed to 75% oxygen at P7-P8 for 16 hours, returned to room air, and euthanized. Retinas from oxygen exposed mice were dissected, fixed, and stained with antiactive Caspase 3.

Immunofluorescence staining of retinal cryosections
Retinas prepared as above were embedded in OCTH, and cut at 6 mm. Sections were stained with rabbit anti-GFAP, rabbit anti-Pax2, or mouse anti-neurofilament (anti-NF, University of Iowa Developmental Studies Hybridoma Bank) in RSB for 2 hours. Afterwards, sections were washed, and stained with appropriate secondary antibodies, including goat anti-rabbit IgG-Alexa fluorH -488, or goat anti-mouse IgG-DyLight 488 (1:200, Jackson ImmunoResearch). After washing and mounting, images were analyzed by confocal microscopy.

Statistical analysis
Data were evaluated by two tailed Student's t-tests using Excel, and are presented as means 6 standard error of means (SEM). All n values refer to number of mice. p,0.05 was considered significant.

Differential but partially overlapping roles of HIF-1a and HIF-2a in retinal vascular development
To compare the roles of HIF-1a and HIF-2a, we generated Hif-1a f/f /Rosa26 CreERT2 (f = floxed) and Hif-2a f/f /Rosa26 CreERT2 mice. To activate CreERT2, neonatal mice were treated with tamoxifen by daily oral gavage at P1-P3. This procedure was previously shown to result in highly efficient deletion of floxed exons in a variety of tissues including the neonatal retina [27,31]. In this study, we were also able to demonstrate significant deletion in retinal tissues ( Figure S1).
To determine if HIF-1a and HIF-2a might have any overlapping functions, we generated Hif-1a f/f /Hif-2a f/f /Rosa26-CreERT2 mice. Vascular development in double disrupted mice was not only defective compared to floxed controls ( Figure 1C, F, I, and L), but was also more severely compromised than in Hif-2a f/f / Rosa26 CreERT2 mice ( Figure 1H, I, K and L). While the number of vascular branches in Hif-2a f/f /Rosa26 CreERT2 mice was about half of floxed controls, the corresponding value in Hif-1a f/f /Hif-2a f/f / Rosa26 CreERT2 mice was further reduced to about 26.0% ( Figure 1M).

Apparently normal retinal vascular development following Tie2 Cre -mediated disruption of floxed Hif-1a or Hif-2a in endothelial cells
Next, we investigated if Hif-1a or Hif-2a disruption in endothelial cells might cause vascular deficiency. For this test, we used Tie2 Cre which was well documented for its high efficiency in deleting floxed DNA sequences in endothelial and hematopoietic cells [26,32]. For example, we showed in a previous study that Tie2 Cre mediated disruption of floxed Vegfr-1 led to vascular defects identical to those in Vegfr-1 germline null embryos [30].
To directly test Cre activity in the context of retinal vascular development, we crossed Tie2 Cre mice with tdTomato reporter mice, the latter of which carried a Cre-inducible tdTomato expression cassette integrated into the Rosa26 locus [25]. Tie2 Cre specifically activated the expression of tdTomato in retinal vascular structures (Figure 2A and B), demonstrating that within the retina, Cre activity was most predominantly located in vascular endothelial cells. In Hif-1a f/f /Tie2 Cre or Hif-2a f/f /Tie2 Cre mice, retinal vascular patterns were indistinguishable from those in floxed mice ( Figure 2C to F), suggesting that endothelial HIF-1a or HIF-2a was not essential for normal vascular development in neonatal retinas.
Reduced development of the astrocytic network in association with HIF-2a deficiency The above findings suggested that the requirement for HIF-a proteins in retinal vascular development might reside in nonendothelial cells. Thus, we examined the development of the retinal astrocytic network by anti-GFAP IF staining in whole mount retinas. At P8, the morphology of astrocytic networks was essentially the same between Hif-1a f/f and Hif-1a f/f /Rosa26 CreERT2 mice ( Figure 3A, D, G, J). However, Hif-2a f/f /Rosa26 CreERT2 mice displayed substantially reduced astrocyte development compared to Hif-2a f/f mice ( Figure 3B, E, H and K). In Hif-2a f/f / Rosa26 CreERT2 mice, retinal tissues were less densely populated with GFAP + astrocytes, and individual astrocyte branches in the network appeared thinner than their counterparts in Hif-2a f/f mice. Interestingly, astrocyte development was further reduced in Figure 3C, F, I, and L).
The development of astrocytic networks was quantified by calculating the percentage of GFAP + tissues in areas midway between the optic nerve and the periphery ( Figure 3M). In Hif-1a f/ f and Hif-1a f/f /Rosa26 CreERT2 mice, GFAP + areas occupied 47.462.8% and 44.963.8% of total areas, respectively (n = 6 per group, p = 0.61). In Hif-2a f/f and Hif-2a f/f /Rosa26 CreERT2 mice, the corresponding values were 46.761.2% and 32.763.9% (n = 7 per group, p,0.05). In Hif-1a f/f /Hif-2a f/f /Rosa26 CreERT2 mice, GFAP + area was down to 21.662.1%, which was not only significantly lower than 45.463.1% in Hif-1a f/f /Hif-2a f/f mice (n = 7 per group, p,0.001), but also lower than in Hif-2a f/f /Rosa26 CreERT2 mice (p,0.05). A to C and G to I are representative images from areas midway between the optic nerve head and periphery; D to F and J to L each displays one of the 4 pedals of flat-mounted retinas. Images were tiled from multiple panels due to the size of retinas. Genotypes at the Hif-a and Rosa26 loci are indicated on top and to the left of the images, respectively. For example, panels G and J are Hif-1a f/f /Rosa26 CreERT2 . All mice were treated with tamoxifen by daily oral gavage at P1 through P3 (40 mg/kg in corn oil). At P8, retinas were dissected and stained with IB 4 -Alexa 594. Note that vascular development was reduced in H and K relative to G and J, respective, and further reduced in I and L. Hyaloid vessels were removed during dissection. Scale bars: A and G, 100 mm; D and J, 500 mm. M. Quantification of vascular branches. LoxP modifications at Hif-1a and Hif-2a loci are indicated below the bar graph, and presence or absence of Rosa26 CreERT2 (R26 CreERT2 ) is indicated by solid or open bars, respectively. Quantification was carried out with the assistance of NIH ImageJ. To average out local variations, vascular branches were counted in 3 rectangular areas (0.260.5 mm 2 ) areas midway between the optic nerve head and the periphery, with an example shown in D. Where a single pedal was not cut wide enough to encompass all three rectangular areas, the third area was counted in an adjacent pedal. Average values from three areas were used as one data point for statistical analysis. n = 6. * p,0.05; ** p,0.01; *** p,0.001. doi:10.1371/journal.pone.0084736.g001

Requirement of astrocyte-derived HIF-2a for vascular development
To determine if HIF-2a was required in astrocytes, we decided to disrupt floxed Hif-2a with GFAP promoter-driven Cre. Several different GFAP Cre lines had been reported previously [23,24,33], two of which were tested in this study, including line 77.6 from the Sofroniew lab and the line from the Messing lab. To examine Cre activity in neonatal retinas, these Cre lines were crossed with tdTomato reporter mice. Unexpectedly, despite reported expression in astrocytes in the brain, line 77.6 failed to activate any tdTomato expression in the retina (data not shown). In contrast, the line from the Messing lab activated tdTomato expression predominantly in Pax2 + astrocyte progenitors and GFAP + astrocytes at P0, P3 and P5 ( Figure S3 and Figure S4). Since we have not investigated GFAP Cre activities at developmental stages beyond the focus of this study, our data do not rule out likely GFAP Cre activity in other cell types in older mice.
We generated Hif-1a f/f /GFAP Cre and Hif-2a f/f /GFAP Cre mice using GFAP Cre mice originated from the Messing lab, and visualized retinal vascular structures at P8 by whole mount staining with IB 4 -Alexa 594. As shown in Figure 4A to C, vascular patterns were similar among Hif-1a f/f , Hif-1a f/f /GFAP Cre and GFAP Cre mice. In sharp contrast, Hif-2a f/f /GFAP Cre mice displayed drastically reduced vascular development relative to floxed controls ( Figure 4D to F). Among 22 Hif-2a f/f /GFAP Cre mice examined at P8, only 2 had remnant amounts of blood vessels as shown in Figure 4D, but the remaining 20 were completely devoid of blood vessels ( Figure 5F). While the primary retinal vascular network was absent in most of Hif-2a f/f /GFAP Cre mice, hyaloid vascular structures were abundantly present ( Figure S5).

Precocious but tapered retinal astrocyte differentiation in
Hif-2a f/f /GFAP Cre mice Data presented above strongly pointed to the possibility that retinal astrocyte development might be defective in Hif-2a f/f / GFAP Cre mice. Unexpectedly, retinas from Hif-2a f/f /GFAP Cre mice exhibited significantly more active astrocyte differentiation at P0 ( Figure 5A, B, I, and J). Strong GFAP + staining was mostly limited to the optic nerve head in floxed mice but extended much further in Hif-2a f/f /GFAP Cre mice. By P3, the development of astrocytic networks started to lag behind in Hif-2a f/f /GFAP Cre mice ( Figure 5C, D, I and J). While strongly GFAP + structures extended similar distances from optic nerves in Hif-2a f/f and Hif-2a f/f / GFAP Cre mice, astrocytes in the latter failed to form complex network organization and occupied less percentage of retinal areas. Further away from the optic nerve head, weakly GFAP + structures were abundantly present in Hif-2a f/f but not Hif-2a f/f / GFAP Cre mice ( Figure 5C and D, upper rectangles and insets).
By P8, Hif-2a f/f /GFAP Cre mice were completely overtaken by floxed controls (Figure 5E to H). Differences in Hif-2a f/f and Hif-2a f/f /GFAP Cre mice were manifested in three ways. First, whereas GFAP + cells had spread out over the entire retinal area in Hif-2a f/f pups, they only reached approximately midway in Hif-2a f/f / GFAP Cre mice ( Figure 5E and F). Second, within GFAP + territories, Hif-2a f/f /GFAP Cre mice were less densely populated with astrocytes ( Figure 5G, H, I and J). Third, astrocytes in Hif-2a f/f /GFAP Cre mice were much thinner and had fewer branches ( Figure 5G and H). In contrast to the dramatic effects of GFAP Cre -mediated Hif-2a disruption on astrocyte development, little difference was found between Hif-1a f/f and Hif-1a f/f /GFAP Cre mice. As shown in Figure  S6A and B, anti-GFAP staining patterns at P8 were indistinguishable between floxed and Hif-1a disrupted mice.
We also evaluated retinal astrocyte development by anti-PDGFRa staining. At P0, PDGFRa positive astrocyte networks expanded towards the retinal periphery by similar distances in Hif-2a f/f and Hif-2a f/f /GFAP Cre mice ( Figure 6A, B and E). Within PDGFRa + territories however, the percentage of area occupied by PDGFRa + structures is much lower in Hif-2a f/f /GFAP Cre mice ( Figure 6A, B and F). At P3, PDGFRa + astrocytes covered essentially the entire retinal inner surface in Hif-2a f/f mice, but were limited to the more central half in Hif-2a f/f /GFAP Cre mice ( Figure 6C, D and E). Within PDGFRa + retinal areas, percentage area occupied by PDGFRa + astrocytes was also significant lower in Hif-2a f/f /GFAP Cre mice ( Figure 6C, D and F).

Depletion of Pax2 + astrocyte progenitors by GFAP Cre -mediated Hif-2a disruption
To further investigate retinal astrocyte differentiation in Hif-2a f/ f /GFAP Cre mice, we analyzed the behavior of Pax2 + cells, which represent retinal astrocyte progenitors. At P0, the presence of Pax2 + cells was significantly reduced in Hif-2a f/f /GFAP Cre retinas ( Figure 7A, B, and K). This difference became more dramatic by P4 ( Figure 7C, D and K). We reasoned that increased astrocyte differentiation at early stages might have occurred at the expense of Pax2 + progenitor population growth, happening too early and too fast, which resulted in the depletion of astrocyte progenitors and immature astrocytes. To test this hypothesis, we directly compared the development of Pax2 + and GFAP + cells by anti-Pax2 and anti-GFAP double IF staining at P0. In Hif-2a f/f mice, GFAP + signals were mostly limited to the optic nerve head area, but Pax2 + cells were abundantly present both near and further beyond the optic nerve head ( Figure 7E and F). Pax2 + and GFAP + domains did not colocalized, but instead formed a complementary pattern ( Figure 7G), demonstrating that Pax2 + population was expanding beyond the optic nerve head without significant astrocyte differentiation at this stage of development. In Hif-2a f/ f /GFAP Cre littermates, retinas had already developed numerous GFAP + sprouts but were less densely populated with Pax2 + cells than in Hif-2a f/f controls ( Figure 7H and I). Merged images also revealed that both Pax2 + and GFAP + cells were present off the optic nerve head in Hif-2a f/f /GFAP Cre mice ( Figure 7J), suggesting that Pax2 + cells were quickly differentiating into GFAP + mature astrocytes.
We also assessed proliferation of Pax2 + cells at P3. While Hif-2a f/f /GFAP Cre mice contained less abundant Pax2 + cells, the percentages of BrdU-labeled Pax2 + cells did not differ between Hif-2a f/f and Hif-2a f/f /GFAP Cre mice ( Figure 8A-I). Apoptotic cells were not detectable by anti-active Caspase 3 staining in both Hif-2a f/f and Hif-2a f/f /GFAP Cre mice ( Figure 8J and K). As a positive control, we performed the same analysis in retinas from hyperoxia treated mice. Apoptotic cells were readily detectable by anti-active Caspase 3 staining in the control ( Figure 8L). Based on these findings, we conclude that loss of Pax2 + cells in Hif-2a f/f /GFAP Cre retinas was unlikely a result of decreased proliferation or increased apoptosis.

Roles of HIF-a proteins in the development of retinal primary vascular plexus
HIF-1a and HIF-2a have well known roles in angiogenesis in a variety of tissues including embryos, ischemic tissues, and tumors [21,[34][35][36][37][38][39][40][41]. In the retina, both HIF-1a and HIF-2a are linked to pathological neoangiogenesis [42][43][44], and HIF-1a in peripheral retinal tissues mediates the development of the intermediate plexus [45]. In this study, global but not endothelial/hematopoietic HIF-2a deficiency led to diminished retinal vascular development. The lack of an essential role for HIF-2a in endothelial cells was not completely unexpected, because mice used in this study were in mixed CD1/B6 background whereas endothelial role of HIF-2a is mostly limited to the 129 strain [35,36,46]. HIF-2a deficiency in the astrocytic lineage led to severe vascular defects. This finding contrasts sharply to normal vascular development a previous study which used a different GFAP Cre line to disrupt Hif-2a [44]. In the study described here, a GFAP Cre line originally generated in the Messing lab was used [23], which robustly activated the expression of Cre-inducible tdTomato reporter in astrocyte progenitors as early as P0. We presume that early expression of GFAP Cre in astrocyte progenitors may be responsible for the phenotypes presented in this manuscript.

Regulation of retinal astrocyte development by HIF-2a
In Hif-2a f/f /GFAP Cre mice, astrocytic templates were only partially developed, and the morphology of HIF-2a deficient astrocytes was abnormally thin and elongated. These defects were associated with almost complete absence of retinal vascular development, suggesting that HIF-2a deficiency compromised not only the abundance but also the function of astrocytes. While the co-existence of vascular and astrocytic defects echoes previous findings that the development of retinal vasculature depends on the astrocytic template [2][3][4], a role of HIF-2a in retinal astrocyte differentiation may have further implications on the relationship between retinal astrocytic and vascular development. Since the stability of HIF-2a is sensitively regulated by prolyl hydroxylase domain proteins in an oxygen dependent manner [47][48][49][50], our findings raise the possibility that regulation of astrocytic and vascular development may be a two way process. In addition to the role of astrocytic templates in regulating retinal vascular growth, their own development may be regulated by oxygen diffusing out of retinal blood vessels. The defects in Hif-2a f/f / GFAP Cre mice may reflect the exaggeration of an inherently important mechanism.
Interpretation of retinal astrocytic phenotypes in Hif-2a f/f / GFAP Cre mice Despite reduced astrocyte abundance in Hif-2a f/f /GFAP Cre retinas after P3, astrocyte differentiation from their progenitors was initially increased instead of being decreased, proliferation index of astrocyte progenitors was not reduced, and apoptosis was not increased. These apparently paradoxical phenomena may be understood by proposing precocious and accelerated differentiation of astrocyte progenitors. Because mature astrocytes are nonproliferative, normal development of the astrocytic network Figure 8. Proliferation and apoptosis assays. A to H. Proliferation assays. Mice were injected with BrdU at P3, and retinas were dissected after an hour. BrdU incorporation in astrocyte progenitors was detected by anti-BrdU and anti-Pax2 double IF staining. A to F show representative images from areas indicated in G and H. Both Pax2 + and BrdU + cells were counted with the assistance of NIH ImageJ program. I. Proliferation index was calculated as % of Pax2 + cells that were also BrdU + . Cre refers to GFAP Cre . n = 4. No significant difference was found between Hif-2a f/f and Hif-2a f/f / GFAP Cre mice. J to L. Apoptosis assay by anti-active Caspase 3 IF staining. J and K are P3 retinas. No apoptotic cells were detected in either Hif-2a f/f or Hif-2a f/f /GFAP Cre mice. L is from mice treated with 75% oxygen for 16 hours between P7 and P8. The same anti-active Caspase 3 staining procedure detected large numbers of apoptotic cells in oxygen-treated mice. Images are representative of data from 4 mice in each group. Scale bars are 50 mm for A to F, 500 mm for G and H, and 50 mm for J to L. doi:10.1371/journal.pone.0084736.g008 depends on the proliferation of astrocyte progenitors and immature astrocytes. If progenitor cells differentiate into mature astrocytes too fast, the supply of the progenitor stock may be depleted because progenitor cells are deprived of the time needed for proliferation. Eventually, dwindling supply of progenitor cells may interrupt astrocyte development. In support of this viewpoint, Hif-2a f/f /GFAP Cre mice indeed contained diminished numbers of Pax2 + astrocyte progenitors as soon as mice were born. Since Pax2 and PDGFRa are co-expressed in immature astrocytes [7], it is not surprising that the number of PDGFRa + cells are similar reduced.
The failure of HIF-2a deficient astrocyte network to expand all the way to retinal periphery might also suggest defects in migration. However, this phenotype can be also explained by accelerated differentiation. As retinal development proceeds, the expansion of mature astrocyte network may depend on in situ differentiation from progenitors seeded at progressively more peripheral positions, rather than migration of fully differentiated astrocytes towards the periphery. Indeed, as early as P3, astrocyte progenitors and immature astrocytes have already spread out to essentially the entire inner surface in wild-type retinas ( Figure S7). In Hif-2a f/f /GFAP Cre mice, the available astrocyte progenitors were enough to cover only the more central part of the retinas, thus precluding astrocyte differentiation at more peripheral positions.

The model
Our data led to a model schematically illustrated in Figure 9, which states that HIF-2a regulates retinal astrocyte development by guarding the rate of astrocyte differentiation from immature progenitors. This function is important because astrocyte progenitors are proliferative whereas mature astrocytes are not. Without HIF-2a, accelerated astrocyte differentiation depletes astrocyte progenitors by disallowing them the time needed for proliferation. As a result, retinal astrocyte development ceases prematurely due to the lack of astrocyte progenitors. It should be pointed out that the physiological HIF-2a level in wild-type retinas is apparently insufficient to completely inhibit astrocyte differentiation, but maintains a balance between progenitor population growth and astrocyte maturation.  Figure S3 GFAP Cre activity in retinal astrocytes. GFAP Cre mice were crossed with transgenic mice carrying a CAG promoter-loxP-Stop-loxP-tdTomato transgene targeted into the ubiquitously expressed Rosa26 locus. A to D. Co-localization of tdTomato expression with GFAP + astrocytes. Retinas were dissected from neonatal mice at P5, and cryosections were cut at 6 mm. Sections were stained with rabbit anti-GFAP and anti-Rabbit IgG-Alexa 488, and analyzed by confocal imaging for tdTomato expression (A) or GFAP + cells (B). Merged images are shown in C and D at different magnifications. tdTomato expression and GFAP + cells colocalized at the inner surface of retinal tissues. E to H. Confocal images of tdTomato expression and anti-NF (neurofilament) immunofluorescence staining (green). Cryosections were prepared as in A to D, but stained with mouse anti-NF followed by goat anti-mouse IgG-DyLight 488. It is evident that tdTomato expression does not colocalize with NF + signals. Scales bars, A-C and E to G, 100 mm; D and H, 50 mm. (TIF) Figure S4 GFAP Cre activity in astrocyte progenitors. GFAP Cre and Cre-inducible tdTomato transgenic mice were crossed, and pups carrying both transgenes were subject to anti-Pax2 or anti-GFAP IF staining at P0 and P3, and visualized by confocal imaging. The vast majority of tdTomato positive cells were also Pax2 + and GFAP + , demonstrating early onset of Cre activity in retinal astrocyte progenitors. n = 3. All images are at the same magnification. Scale bar represents 50 mm. (TIF) Figure S5 Hyaloid vessels in Hif-2a f/f /GFAP Cre mice. Retinas were dissected at P8, care being taken to preserve hyaloid vessels. A and B. IB 4 -Alexa 594-stained retinas. C and D. Higher magnifications of the top right quarter from A and B, respectively. In C, white arrows point to main arterioles; green arrows indicate large branches between arterioles. Since every main arteriole has just one accompanying venules in normal retinas, additional large branches may be hyaloid vessels. Hyaloid vessels are more prominently present in Hif-2a f/f /GFAP Cre mice, presumably to compensate for the loss of retinal blood vessels. n = 3. (TIF) Figure S6 Apparently normal astrocyte development in Hif-1a f/f /GFAP Cre mice. At P8, retinas were dissected from Hif-1a f/f (A) and Hif-1a f/f /GFAP Cre (B) mice, fixed, and stained by rabbit anti-GFAP followed by goat anti-rabbit IgG-Alexa 488. Figure 9. Model for HIF-2a regulated retinal astrocyte differentiation. In wild-type mice, physiological level of HIF-2a may partially suppress differentiation of mature astrocytes (green stars) from their progenitors (orange ovals). Thus, while a proportion of astrocyte progenitors differentiate into non-proliferative mature astrocytes, the rest may undergo active proliferation to replenish progenitor populations. HIF-2a deficiency disturbs the balance between progenitor proliferation and astrocyte differentiation. Accelerated astrocyte differentiation may cause rapid loss of astrocyte progenitors due to inadequate time for proliferation, leading to unsustainable astrocyte development. doi:10.1371/journal.pone.0084736.g009

Supporting Information
Data shown are representative confocal images from 4 mice in each genotype. Scales bars are 100 mm. Images are representative of data from 3 mice per group. (TIF) Figure S7 Astrocyte progenitors and immature astrocytes in P3 retinas. A to C are confocal images of wild-type (Hif-2a f/f ) retinas stained with indicated antibodies at P3. Boxed areas are expanded and shown in D to F. Images in D to F were brightened to show more details at the periphery. These data confirm that by P3, astrocyte progenitors and immature astrocytes are already present throughout the retina. Images are represen-tative of at least 3 mice each. Scales bars are 500 mm for A-C, and 100 mm for D to F. (TIF)

Author Contributions
Conceived and designed the experiments: GHF. Performed the experiments: LJD KT. Analyzed the data: LJD KT GHF. Contributed reagents/ materials/analysis tools: LJD KT GHF. Wrote the paper: LJD GHF.